Design of machine learning-based water quality prediction system with recursive feature elimination cross-validation

Abstract

Lack of clean water has become a problem in the world, and it is estimated that by 2025 there will be 2.8 billion people who will experience a shortage of clean water. The high demand for clean water and the limited water sources with proper potency is one of the main reasons for the need for a device capable of measuring the potability level of water that is flexible to carry and does not require high costs in the manufacturing process. In this paper, the design of machine learning-based potability devices with recursive feature elimination with cross-validation (RFECV) is carried out as a guide in making the design of a water potability detection system, and the results obtained from RFECV with the Random Forest (RF) algorithm have a higher accuracy value. 15.71% better than the RF model, 6.85% better than the Support Vector Machine (SVM) model, and 8.57% better than the Artificial Neural Network (ANN) model trained without RFECV. The water potability prediction system's design selection is based on feature elimination results in the RFECV process. It is based on a literature review on device selection. The proposed water potability detection system consists of ESP32 as the primary computing device, electrochemical spectroscopy-based Al/PET sensor to detect sulfate values with a sensitivity of 0.874 Ω/ppm, PH4502C as a pH measuring instrument with an accuracy of up to 98.10%, WD-35802-49 electrode. as a device for measuring hardness in water with a measurement range of 0.4 – 40,000 ppm, a total dissolved solids sensor to determine the solids content in water with an accuracy of up to 97.80%, as well as a carbon-based sensor for measuring chloramines with a reading capacity of 186 nA/ppm.